Techniques for Dealing with Missing Values in Feedforward Networks
|
|
- Nathan Hutchinson
- 5 years ago
- Views:
Transcription
1 Techniques for Dealing with Missing Values in Feedforward Networks Peter Vamplew, David Clark*, Anthony Adams, Jason Muench Artificial Neural Networks Research Group, Department of Computer Science, University of Tasmania * permanent address: Department of Information Science and Engineering, University of Canberra Abstract: Missing or incomplete data, a common reality, causes problems for artificial neural networks. In this paper we investigate several methods for dealing with missing values in feedforward networks. Reduced networks, substitution, estimation and expanded networks are applied to three data sets. We find that data sets vary in their sensitivity to missing values, and that reduced networks and estimation are the most effective ways of dealing with them. Introduction Artificial neural networks trained using backpropagation have been used for a wide variety of classification problems. In many real world problems, however, some of the data may be missing or incomplete. This causes particular problems for artificial neural networks as the distributed nature of the processing makes it very difficult to isolate effects due to one variable. The aim of this study is to compare different techniques for dealing with missing data. We assume a complete set of training data is available and a single hidden layer backpropagation network. This work builds on and extends earlier work of Vamplew and Adams 1. Terminology In this paper we will adopt the following terminology: V = number of variables; N, H, M = number of input, hidden, output nodes; and µ, σ = mean, standard deviation (of the training data) Description of the data We have applied the techniques to three data sets, the Iris, the Weedseed and the Handshape data set. All data are normalised to the range ±1. The Iris data set 2 has been used widely as a testbed for statistical analysis techniques. The sepal length, sepal width, petal length, and petal width were measured on 50 iris specimens from each of 3 species, Iris setosa, Iris versicolor, and Iris virginica. There are thus 150 pieces of data, of which 100 are used for training and 50 for testing. The proportion of data which is correctly classified is high. In the Weedseed data set 3, weed seeds are classified into one of ten types, based on seven measurements of dimensions of the seeds. The seed data consists of measurements of 398 different seeds, giving 40 examples of each seed type, apart from two types which have only 39 examples. A test set consisting of ten examples of each seed type is extracted from these data, and the remaining 298 examples are used as the training set. The proportion of data which is correctly classified is only moderate. The Handshape data set 4 was developed by measuring the values returned by 16 sensors on various hand and finger joints when the hand was positioned in 20 different hand shapes used in the Auslan sign language. A training set of 2000 examples and a test set of 200 examples were created by adding noise to the 20 original examples. This data set contains many redundant inputs and a high proportion of the data can be classified correctly even with high levels of added noise. Thus for the purposes of this study, we used only 7 variables for each data point. Experimental details The networks are trained using the standard pattern presentation backpropagation algorithm which updates the weights after each
2 (randomly-chosen) presentation of the data. We use a fixed number of presentations and a fixed number of hidden nodes for each data set. In each trial, there are 10 runs, each with different starting weights. For all training the training rate was 0.1 and the momentum 0.
3 Data set Presentatio ns Architecture (N H M) # training points # testing points Iris 50, Weedseed 100, Handshape 100, Methods We investigated four general approaches for dealing with missing values: Reduced network - A separate network is trained for each missing value, each having V-1 nodes in the input layer; Substitution - Another value is substituted for the missing value; Estimation - A value for the missing value is estimated from the remaining data. This is substituted for the missing value; Expanded network - The standard network architecture is modified by the addition of extra input nodes in such a manner as to allow the network to distinguish between missing and complete data. Reduced networks V networks are trained, each with one variable absent (N = V-1). As each network is trained on all of the data available, and is of the right size for the data, it is expected to perform as well as possible. It should give an upper bound on what is possible and provide a baseline for judging other methods. A major disadvantage of this approach is that it requires N+1 networks to cater for all possible cases of one missing value. If it is to be applied to data points with two missing values, it will require an additional N 2 / 2 networks. Substitution Here a constant value for substitution is found for each variable that might be missing. This is then substituted in the complete network (ie. the network trained on complete data). We report on substituting the mean, median and zero (the mid-range). Other substitutions including the minimum, the maximum, committeeing the results from a series of values, and a random value were tried, but all gave poorer results and are not further reported. The mean is calculated for each variable over all examples in the training set. It is fixed for all data points, that is the mean of the variable in the training set is used whenever that variable is missing. It is a very simple approach, easy to understand and requires no extra networks or training. The median and mid-ranges are treated similarly. Estimation The best estimate is derived by using a separate network to estimate the missing value from the values that are present. Thus there is only one classification network, supported by V estimating networks (each with N = V-1). Unlike the classifier networks, the supporting networks have a single analogue output. This approach relies on the input values being correlated, which is true for most real data sets. Expanded networks The aim here is to try to let the network know whether data is complete or has missing values. This is achieved by using two input nodes for each input variable. Three types of expanded networks are investigated, flagged, high/low and shadow weight networks. In flagged networks, each data value is associated with a value node and a flag node. If the value is known, it is input to the value node and 0 is input to the flag node. If the value is missing, that attribute s mean is input to the value node and 1 is input to the flag node.
4 In high/low networks, each data value is associated with a high node and a low node. If the value is known, it is input to both the low and high nodes. If the value is missing, predetermined constants are input to the low and high nodes. Thus this technique essentially treats missing data as fuzzy. Three sets of constants are tried: low = minimum value, high = maximum value; low = µ σ, high = µ + σ ; and low = µ 2 σ, high = µ +2 σ. In the shadow weight networks, each data value is associated with a standard node and a shadow node. If the value is known, it is input to the standard node and 0 is input to the shadow node. If the value is missing, 0 is input to the standard node, and 1 is input to the shadow node. The main difference between the shadowed network and other expanded networks is in the training. This occurs in two phases. Firstly, a complete network (with no weights from shadow nodes to hidden layer nodes) is trained using complete data. Then the weights of the links between the shadow nodes and nodes in the hidden layer are trained. The aim is to create a network in which the performance of complete data is not degraded. The performance improves when several shadow hidden nodes are added to the hidden layer. These nodes are completely connected to the input layer, but are only trained during the second phase of training and only used when the input data has missing values. All of the expanded networks require extra training data. The training data is augmented with data where there have been substitutions for (single) values (that is, one value per data point). This means that there will be (V+1) times as much training data. Figure 1: A shadow weights network. The bold lines indicate standard weights, the dashed lines and shaded circle represent shadow weights and node (in this case, the lowest input value is unknown) Results Results are set out in the tables below, shaded cells indicating a significant reduction in performance (using the reduced networks as a baseline). They show that only reduced networks and estimation are effective on all data sets. The results also show a significant difference between data sets. For the Iris data, all techniques work well. For the Weedseed data, only reduced networks and estimation work well. For the Handshape data, all techniques worked well except substitution. A tentative conclusion here is that data which is hard to classify is more susceptible to missing values and needs more work if data points with missing values are to be useful. Iris data Variable missing none Reduced network Substitute mean Substitute median Substitute zero Estimation
5 Flagged (min max) ( µ ± σ ) ( µ ±2 σ ) Shadow weights Weedseed data Variable missing none Reduced network Substitute (mean) Substitute (median) Substitute (zero) Estimation Flagged (min max) ( µ ± σ ) ( µ ±2 σ ) Shadowed weights Handshape data Variable missing none Reduced network Substitute (mean) Substitute (median) Substitute (zero) Estimation Flagged
6 (min max) ( µ ± σ ) ( µ ±2 σ ) Shadowed weights Equivalence of networks Several of the networks described above are topographically equivalent, a simple transformation turning a network into another of different type which will give exactly the same output for equivalent data. There exist 1-1 mappings between all expanded networks, except for the shadow network with additional hidden layer nodes. Moreover, there is a mapping (not 1-1) from the complete network to each expanded network. Thus all expanded networks should perform equivalently, and they should all perform at least as well as substitution. The results are in broad agreement with this analysis. Multiple Missing Values The different missing values techniques vary in terms of the ease with which they can be scaled to handle examples with multiple missing inputs. Substitution and expanded networks cope readily with this situation as all that is required is to input the appropriate value(s) for each missing input value. The main issue is whether these networks can generalise from the examples with single missing values seen during training, or whether it is necessary to also train the networks on examples with multiple missing values. Estimation and reduced networks do not scale well to multiple missing values as they suffer from a combinatorial expansion in the number of networks to be trained. It is possible to adapt estimation to the case of multiple missing values however, by combining it with the substitution techniques. The estimate networks are trained and used to produce an estimate of each missing input as before, with the exception that substitution is performed for any missing values which are required as inputs to these estimate networks. The estimates produced in this way can then be used as inputs for the main classification network. Vamplew 5 has shown that expanded networks and substitution generalise poorly with multiple missing values if trained using single missing values. He has also shown that the combines substitution/estimation technique is effective even when training is performed on single missing values. Conclusions 1. All techniques perform well on complete data. Thus techniques such as shadow networks which are designed specifically to ensure that the performance on complete data does not degrade are not necessary. 2. Techniques involving fixed valued substitution for missing values do not perform as well as ones which estimate the missing values from the non-missing values for that data point. 3. Data sets vary in their sensitivity to missing values. If the data set is insensitive to missing values, all techniques are effective, but if it is sensitive to missing values, the choice of technique is important. References 1 Vamplew, P and Adams, T., Missing Values in a Backpropogation Neural Net, in Leong, S and Jabri, M (eds)
7 Proceedings of the Third Australian Conference on Neural Networks, Sydney, Feb, 1992, Fisher, R.A., The use of multiple measurements in Taxonomic Problems, Annals of Eugenics, 7, 1936, Weedseed data, obtained from University of Stirling, in private communication, Collier, P.A., Department of Computer Science, University of Tasmamia. 4 Vamplew, P. and Adams, A, The SLARTI System: Applying Artificial Neural Networks to Sign Language Recognition in Proceedings of the Conference on Technology and Persons with Disabilities, California State University, Northridge, March, Vamplew, P., Computer Recognition of Hand Gestures, Ph. D. thesis, Department of Computer Science, University of Tasmania, forthcoming.
Introduction to Artificial Intelligence
Introduction to Artificial Intelligence COMP307 Machine Learning 2: 3-K Techniques Yi Mei yi.mei@ecs.vuw.ac.nz 1 Outline K-Nearest Neighbour method Classification (Supervised learning) Basic NN (1-NN)
More informationNeural Networks Laboratory EE 329 A
Neural Networks Laboratory EE 329 A Introduction: Artificial Neural Networks (ANN) are widely used to approximate complex systems that are difficult to model using conventional modeling techniques such
More informationSimulation of Back Propagation Neural Network for Iris Flower Classification
American Journal of Engineering Research (AJER) e-issn: 2320-0847 p-issn : 2320-0936 Volume-6, Issue-1, pp-200-205 www.ajer.org Research Paper Open Access Simulation of Back Propagation Neural Network
More informationKTH ROYAL INSTITUTE OF TECHNOLOGY. Lecture 14 Machine Learning. K-means, knn
KTH ROYAL INSTITUTE OF TECHNOLOGY Lecture 14 Machine Learning. K-means, knn Contents K-means clustering K-Nearest Neighbour Power Systems Analysis An automated learning approach Understanding states in
More informationData Mining: Exploring Data. Lecture Notes for Chapter 3
Data Mining: Exploring Data Lecture Notes for Chapter 3 Slides by Tan, Steinbach, Kumar adapted by Michael Hahsler Look for accompanying R code on the course web site. Topics Exploratory Data Analysis
More informationAn Efficient Method for Extracting Fuzzy Classification Rules from High Dimensional Data
Published in J. Advanced Computational Intelligence, Vol., No., 997 An Efficient Method for Extracting Fuzzy Classification Rules from High Dimensional Data Stephen L. Chiu Rockwell Science Center 049
More informationCluster Analysis using Spherical SOM
Cluster Analysis using Spherical SOM H. Tokutaka 1, P.K. Kihato 2, K. Fujimura 2 and M. Ohkita 2 1) SOM Japan Co-LTD, 2) Electrical and Electronic Department, Tottori University Email: {tokutaka@somj.com,
More informationClassification with Diffuse or Incomplete Information
Classification with Diffuse or Incomplete Information AMAURY CABALLERO, KANG YEN Florida International University Abstract. In many different fields like finance, business, pattern recognition, communication
More informationA New Approach for Handling the Iris Data Classification Problem
International Journal of Applied Science and Engineering 2005. 3, : 37-49 A New Approach for Handling the Iris Data Classification Problem Shyi-Ming Chen a and Yao-De Fang b a Department of Computer Science
More informationInput: Concepts, Instances, Attributes
Input: Concepts, Instances, Attributes 1 Terminology Components of the input: Concepts: kinds of things that can be learned aim: intelligible and operational concept description Instances: the individual,
More informationMATH5745 Multivariate Methods Lecture 13
MATH5745 Multivariate Methods Lecture 13 April 24, 2018 MATH5745 Multivariate Methods Lecture 13 April 24, 2018 1 / 33 Cluster analysis. Example: Fisher iris data Fisher (1936) 1 iris data consists of
More informationAn Introduction to Cluster Analysis. Zhaoxia Yu Department of Statistics Vice Chair of Undergraduate Affairs
An Introduction to Cluster Analysis Zhaoxia Yu Department of Statistics Vice Chair of Undergraduate Affairs zhaoxia@ics.uci.edu 1 What can you say about the figure? signal C 0.0 0.5 1.0 1500 subjects Two
More informationRadial Basis Function (RBF) Neural Networks Based on the Triple Modular Redundancy Technology (TMR)
Radial Basis Function (RBF) Neural Networks Based on the Triple Modular Redundancy Technology (TMR) Yaobin Qin qinxx143@umn.edu Supervisor: Pro.lilja Department of Electrical and Computer Engineering Abstract
More informationCOMPARISON OF DENSITY-BASED CLUSTERING ALGORITHMS
COMPARISON OF DENSITY-BASED CLUSTERING ALGORITHMS Mariam Rehman Lahore College for Women University Lahore, Pakistan mariam.rehman321@gmail.com Syed Atif Mehdi University of Management and Technology Lahore,
More informationStats 170A: Project in Data Science Exploratory Data Analysis: Clustering Algorithms
Stats 170A: Project in Data Science Exploratory Data Analysis: Clustering Algorithms Padhraic Smyth Department of Computer Science Bren School of Information and Computer Sciences University of California,
More informationData Mining: Exploring Data. Lecture Notes for Chapter 3
Data Mining: Exploring Data Lecture Notes for Chapter 3 1 What is data exploration? A preliminary exploration of the data to better understand its characteristics. Key motivations of data exploration include
More informationExperimental Design + k- Nearest Neighbors
10-601 Introduction to Machine Learning Machine Learning Department School of Computer Science Carnegie Mellon University Experimental Design + k- Nearest Neighbors KNN Readings: Mitchell 8.2 HTF 13.3
More informationApplication of Fuzzy Logic Akira Imada Brest State Technical University
A slide show of our Lecture Note Application of Fuzzy Logic Akira Imada Brest State Technical University Last modified on 29 October 2016 (Contemporary Intelligent Information Techniques) 2 I. Fuzzy Basic
More informationNearest Neighbor Classification
Nearest Neighbor Classification Professor Ameet Talwalkar Professor Ameet Talwalkar CS260 Machine Learning Algorithms January 11, 2017 1 / 48 Outline 1 Administration 2 First learning algorithm: Nearest
More informationClassification with Diffuse or Incomplete Information
Classification with Diffuse or Incomplete Information AMAURY CABALLERO, KANG YEN, YECHANG FANG Department of Electrical & Computer Engineering Florida International University 10555 W. Flagler Street,
More informationData Mining: Exploring Data. Lecture Notes for Chapter 3. Introduction to Data Mining
Data Mining: Exploring Data Lecture Notes for Chapter 3 Introduction to Data Mining by Tan, Steinbach, Kumar What is data exploration? A preliminary exploration of the data to better understand its characteristics.
More informationPerformance Analysis of Data Mining Classification Techniques
Performance Analysis of Data Mining Classification Techniques Tejas Mehta 1, Dr. Dhaval Kathiriya 2 Ph.D. Student, School of Computer Science, Dr. Babasaheb Ambedkar Open University, Gujarat, India 1 Principal
More informationIN recent years, neural networks have attracted considerable attention
Multilayer Perceptron: Architecture Optimization and Training Hassan Ramchoun, Mohammed Amine Janati Idrissi, Youssef Ghanou, Mohamed Ettaouil Modeling and Scientific Computing Laboratory, Faculty of Science
More informationChuck Cartledge, PhD. 20 January 2018
Big Data: Data Analysis Boot Camp Visualizing the Iris Dataset Chuck Cartledge, PhD 20 January 2018 1/31 Table of contents (1 of 1) 1 Intro. 2 Histograms Background 3 Scatter plots 4 Box plots 5 Outliers
More informationANN exercise session
ANN exercise session In this exercise session, you will read an external file with Iris flowers and create an internal database in Java as it was done in previous exercise session. A new file contains
More informationApplication of Or-based Rule Antecedent Fuzzy Neural Networks to Iris Data Classification Problem
Vol.1 (DTA 016, pp.17-1 http://dx.doi.org/10.157/astl.016.1.03 Application of Or-based Rule Antecedent Fuzzy eural etworks to Iris Data Classification roblem Chang-Wook Han Department of Electrical Engineering,
More informationData Mining - Data. Dr. Jean-Michel RICHER Dr. Jean-Michel RICHER Data Mining - Data 1 / 47
Data Mining - Data Dr. Jean-Michel RICHER 2018 jean-michel.richer@univ-angers.fr Dr. Jean-Michel RICHER Data Mining - Data 1 / 47 Outline 1. Introduction 2. Data preprocessing 3. CPA with R 4. Exercise
More informationWhy MultiLayer Perceptron/Neural Network? Objective: Attributes:
Why MultiLayer Perceptron/Neural Network? Neural networks, with their remarkable ability to derive meaning from complicated or imprecise data, can be used to extract patterns and detect trends that are
More informationFINDING PATTERN BEHAVIOR IN TEMPORAL DATA USING FUZZY CLUSTERING
1 FINDING PATTERN BEHAVIOR IN TEMPORAL DATA USING FUZZY CLUSTERING RICHARD E. HASKELL Oaland University PING LI Oaland University DARRIN M. HANNA Oaland University KA C. CHEOK Oaland University GREG HUDAS
More informationCROSS-CORRELATION NEURAL NETWORK: A NEW NEURAL NETWORK CLASSIFIER
CROSS-CORRELATION NEURAL NETWORK: A NEW NEURAL NETWORK CLASSIFIER ARIT THAMMANO* AND NARODOM KLOMIAM** Faculty of Information Technology King Mongkut s Institute of Technology Ladkrang, Bangkok, 10520
More informationModel Selection Introduction to Machine Learning. Matt Gormley Lecture 4 January 29, 2018
10-601 Introduction to Machine Learning Machine Learning Department School of Computer Science Carnegie Mellon University Model Selection Matt Gormley Lecture 4 January 29, 2018 1 Q&A Q: How do we deal
More informationCS 584 Data Mining. Classification 1
CS 584 Data Mining Classification 1 Classification: Definition Given a collection of records (training set ) Each record contains a set of attributes, one of the attributes is the class. Find a model for
More informationMachine Learning 13. week
Machine Learning 13. week Deep Learning Convolutional Neural Network Recurrent Neural Network 1 Why Deep Learning is so Popular? 1. Increase in the amount of data Thanks to the Internet, huge amount of
More informationData Mining: Exploring Data. Lecture Notes for Data Exploration Chapter. Introduction to Data Mining
Data Mining: Exploring Data Lecture Notes for Data Exploration Chapter Introduction to Data Mining by Tan, Steinbach, Karpatne, Kumar 02/03/2018 Introduction to Data Mining 1 What is data exploration?
More informationk Nearest Neighbors Super simple idea! Instance-based learning as opposed to model-based (no pre-processing)
k Nearest Neighbors k Nearest Neighbors To classify an observation: Look at the labels of some number, say k, of neighboring observations. The observation is then classified based on its nearest neighbors
More informationInstantaneously trained neural networks with complex inputs
Louisiana State University LSU Digital Commons LSU Master's Theses Graduate School 2003 Instantaneously trained neural networks with complex inputs Pritam Rajagopal Louisiana State University and Agricultural
More informationA Systematic Overview of Data Mining Algorithms. Sargur Srihari University at Buffalo The State University of New York
A Systematic Overview of Data Mining Algorithms Sargur Srihari University at Buffalo The State University of New York 1 Topics Data Mining Algorithm Definition Example of CART Classification Iris, Wine
More informationHands on Datamining & Machine Learning with Weka
Step1: Click the Experimenter button to launch the Weka Experimenter. The Weka Experimenter allows you to design your own experiments of running algorithms on datasets, run the experiments and analyze
More informationA Class of Instantaneously Trained Neural Networks
A Class of Instantaneously Trained Neural Networks Subhash Kak Department of Electrical & Computer Engineering, Louisiana State University, Baton Rouge, LA 70803-5901 May 7, 2002 Abstract This paper presents
More informationA Classifier with the Function-based Decision Tree
A Classifier with the Function-based Decision Tree Been-Chian Chien and Jung-Yi Lin Institute of Information Engineering I-Shou University, Kaohsiung 84008, Taiwan, R.O.C E-mail: cbc@isu.edu.tw, m893310m@isu.edu.tw
More informationNeural Network and Adaptive Feature Extraction Technique for Pattern Recognition
Material Science Research India Vol. 8(1), 01-06 (2011) Neural Network and Adaptive Feature Extraction Technique for Pattern Recognition WAEL M. KHEDR and QAMAR A. AWAD Mathematical Department, Faculty
More informationData Mining: Exploring Data
Data Mining: Exploring Data Lecture Notes for Chapter 3 Introduction to Data Mining by Tan, Steinbach, Kumar But we start with a brief discussion of the Friedman article and the relationship between Data
More informationCluster Analysis and Visualization. Workshop on Statistics and Machine Learning 2004/2/6
Cluster Analysis and Visualization Workshop on Statistics and Machine Learning 2004/2/6 Outlines Introduction Stages in Clustering Clustering Analysis and Visualization One/two-dimensional Data Histogram,
More informationIEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 10, NO. 6, NOVEMBER Inverting Feedforward Neural Networks Using Linear and Nonlinear Programming
IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 10, NO. 6, NOVEMBER 1999 1271 Inverting Feedforward Neural Networks Using Linear and Nonlinear Programming Bao-Liang Lu, Member, IEEE, Hajime Kita, and Yoshikazu
More informationStatistical Methods in AI
Statistical Methods in AI Distance Based and Linear Classifiers Shrenik Lad, 200901097 INTRODUCTION : The aim of the project was to understand different types of classification algorithms by implementing
More informationCSCI567 Machine Learning (Fall 2014)
CSCI567 Machine Learning (Fall 2014) Drs. Sha & Liu {feisha,yanliu.cs}@usc.edu September 9, 2014 Drs. Sha & Liu ({feisha,yanliu.cs}@usc.edu) CSCI567 Machine Learning (Fall 2014) September 9, 2014 1 / 47
More informationSummary. Machine Learning: Introduction. Marcin Sydow
Outline of this Lecture Data Motivation for Data Mining and Learning Idea of Learning Decision Table: Cases and Attributes Supervised and Unsupervised Learning Classication and Regression Examples Data:
More informationWeight Optimization for a Neural Network using Particle Swarm Optimization (PSO)
Institute of Integrated Sensor Systems Dept. of Electrical Engineering and Information Technology Weight Optimization for a Neural Network using Particle Swarm Optimization (PSO) Stefanie Peters October
More information2002 Journal of Software.. (stacking).
1000-9825/2002/13(02)0245-05 2002 Journal of Software Vol13, No2,,, (,200433) E-mail: {wyji,ayzhou,zhangl}@fudaneducn http://wwwcsfudaneducn : (GA) (stacking), 2,,, : ; ; ; ; : TP18 :A, [1],,, :,, :,,,,
More information2. Basic Task of Pattern Classification
2. Basic Task of Pattern Classification Definition of the Task Informal Definition: Telling things apart 3 Definition: http://www.webopedia.com/term/p/pattern_recognition.html pattern recognition Last
More informationEFFECTIVENESS PREDICTION OF MEMORY BASED CLASSIFIERS FOR THE CLASSIFICATION OF MULTIVARIATE DATA SET
EFFECTIVENESS PREDICTION OF MEMORY BASED CLASSIFIERS FOR THE CLASSIFICATION OF MULTIVARIATE DATA SET C. Lakshmi Devasena 1 1 Department of Computer Science and Engineering, Sphoorthy Engineering College,
More informationRepresenting structural patterns: Reading Material: Chapter 3 of the textbook by Witten
Representing structural patterns: Plain Classification rules Decision Tree Rules with exceptions Relational solution Tree for Numerical Prediction Instance-based presentation Reading Material: Chapter
More informationA Generalized Method to Solve Text-Based CAPTCHAs
A Generalized Method to Solve Text-Based CAPTCHAs Jason Ma, Bilal Badaoui, Emile Chamoun December 11, 2009 1 Abstract We present work in progress on the automated solving of text-based CAPTCHAs. Our method
More informationk-nearest Neighbors + Model Selection
10-601 Introduction to Machine Learning Machine Learning Department School of Computer Science Carnegie Mellon University k-nearest Neighbors + Model Selection Matt Gormley Lecture 5 Jan. 30, 2019 1 Reminders
More informationImage processing & Computer vision Xử lí ảnh và thị giác máy tính
Image processing & Computer vision Xử lí ảnh và thị giác máy tính Detection and Recognition 2D et 3D Alain Boucher - IFI Introduction In this chapter, we introduce some techniques for pattern detection
More informationThe Potential of Prototype Styles of Generalization. D. Randall Wilson Tony R. Martinez
Proceedings of the 6th Australian Joint Conference on Artificial Intelligence (AI 93), pp. 356-361, Nov. 1993. The Potential of Prototype Styles of Generalization D. Randall Wilson Tony R. Martinez Computer
More informationHsiaochun Hsu Date: 12/12/15. Support Vector Machine With Data Reduction
Support Vector Machine With Data Reduction 1 Table of Contents Summary... 3 1. Introduction of Support Vector Machines... 3 1.1 Brief Introduction of Support Vector Machines... 3 1.2 SVM Simple Experiment...
More informationCombined Weak Classifiers
Combined Weak Classifiers Chuanyi Ji and Sheng Ma Department of Electrical, Computer and System Engineering Rensselaer Polytechnic Institute, Troy, NY 12180 chuanyi@ecse.rpi.edu, shengm@ecse.rpi.edu Abstract
More informationConstructing Hidden Units using Examples and Queries
Constructing Hidden Units using Examples and Queries Eric B. Baum Kevin J. Lang NEC Research Institute 4 Independence Way Princeton, NJ 08540 ABSTRACT While the network loading problem for 2-layer threshold
More informationArtificial Neural Network based Curve Prediction
Artificial Neural Network based Curve Prediction LECTURE COURSE: AUSGEWÄHLTE OPTIMIERUNGSVERFAHREN FÜR INGENIEURE SUPERVISOR: PROF. CHRISTIAN HAFNER STUDENTS: ANTHONY HSIAO, MICHAEL BOESCH Abstract We
More informationPerformance Measure of Hard c-means,fuzzy c-means and Alternative c-means Algorithms
Performance Measure of Hard c-means,fuzzy c-means and Alternative c-means Algorithms Binoda Nand Prasad*, Mohit Rathore**, Geeta Gupta***, Tarandeep Singh**** *Guru Gobind Singh Indraprastha University,
More information2009 Inderscience Enterprises. Reprinted with permission.
Xiaolei Wang, Xiao Zhi Gao, and Seppo J. Ovaska. 29. Fusion of clonal selection algorithm and harmony search method in optimisation of fuzzy classification systems. International Journal of Bio Inspired
More informationValue Difference Metrics for Continuously Valued Attributes
Proceedings of the International Conference on Artificial Intelligence, Expert Systems and Neural Networks (AIE 96), pp. 11-14, 1996. Value Difference Metrics for Continuously Valued Attributes D. Randall
More informationChapter 4. The Classification of Species and Colors of Finished Wooden Parts Using RBFNs
Chapter 4. The Classification of Species and Colors of Finished Wooden Parts Using RBFNs 4.1 Introduction In Chapter 1, an introduction was given to the species and color classification problem of kitchen
More informationRobustness of Selective Desensitization Perceptron Against Irrelevant and Partially Relevant Features in Pattern Classification
Robustness of Selective Desensitization Perceptron Against Irrelevant and Partially Relevant Features in Pattern Classification Tomohiro Tanno, Kazumasa Horie, Jun Izawa, and Masahiko Morita University
More information3 Virtual attribute subsetting
3 Virtual attribute subsetting Portions of this chapter were previously presented at the 19 th Australian Joint Conference on Artificial Intelligence (Horton et al., 2006). Virtual attribute subsetting
More informationImproving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah
Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Reference Most of the slides are taken from the third chapter of the online book by Michael Nielson: neuralnetworksanddeeplearning.com
More informationMachine Learning for. Artem Lind & Aleskandr Tkachenko
Machine Learning for Object Recognition Artem Lind & Aleskandr Tkachenko Outline Problem overview Classification demo Examples of learning algorithms Probabilistic modeling Bayes classifier Maximum margin
More informationSubgoal Chaining and the Local Minimum Problem
Subgoal Chaining and the Local Minimum roblem Jonathan. Lewis (jonl@dcs.st-and.ac.uk), Michael K. Weir (mkw@dcs.st-and.ac.uk) Department of Computer Science, University of St. Andrews, St. Andrews, Fife
More informationMachine Learning Chapter 2. Input
Machine Learning Chapter 2. Input 2 Input: Concepts, instances, attributes Terminology What s a concept? Classification, association, clustering, numeric prediction What s in an example? Relations, flat
More informationTransactions on Information and Communications Technologies vol 16, 1996 WIT Press, ISSN
Comparative study of fuzzy logic and neural network methods in modeling of simulated steady-state data M. Järvensivu and V. Kanninen Laboratory of Process Control, Department of Chemical Engineering, Helsinki
More informationOrdering attributes for missing values prediction and data classification
Ordering attributes for missing values prediction and data classification E. R. Hruschka Jr., N. F. F. Ebecken COPPE /Federal University of Rio de Janeiro, Brazil. Abstract This work shows the application
More informationData Mining. Practical Machine Learning Tools and Techniques. Slides for Chapter 3 of Data Mining by I. H. Witten, E. Frank and M. A.
Data Mining Practical Machine Learning Tools and Techniques Slides for Chapter 3 of Data Mining by I. H. Witten, E. Frank and M. A. Hall Output: Knowledge representation Tables Linear models Trees Rules
More informationBoxplot
Boxplot By: Meaghan Petix, Samia Porto & Franco Porto A boxplot is a convenient way of graphically depicting groups of numerical data through their five number summaries: the smallest observation (sample
More informationClassification: Decision Trees
Classification: Decision Trees IST557 Data Mining: Techniques and Applications Jessie Li, Penn State University 1 Decision Tree Example Will a pa)ent have high-risk based on the ini)al 24-hour observa)on?
More informationArtificial Neuron Modelling Based on Wave Shape
Artificial Neuron Modelling Based on Wave Shape Kieran Greer, Distributed Computing Systems, Belfast, UK. http://distributedcomputingsystems.co.uk Version 1.2 Abstract This paper describes a new model
More informationIntroduction to digital image classification
Introduction to digital image classification Dr. Norman Kerle, Wan Bakx MSc a.o. INTERNATIONAL INSTITUTE FOR GEO-INFORMATION SCIENCE AND EARTH OBSERVATION Purpose of lecture Main lecture topics Review
More informationUSING REGRESSION TREES IN PREDICTIVE MODELLING
Production Systems and Information Engineering Volume 4 (2006), pp. 115-124 115 USING REGRESSION TREES IN PREDICTIVE MODELLING TAMÁS FEHÉR University of Miskolc, Hungary Department of Information Engineering
More informationAn Approach to Identify the Number of Clusters
An Approach to Identify the Number of Clusters Katelyn Gao Heather Hardeman Edward Lim Cristian Potter Carl Meyer Ralph Abbey July 11, 212 Abstract In this technological age, vast amounts of data are generated.
More informationLinear discriminant analysis and logistic
Practical 6: classifiers Linear discriminant analysis and logistic This practical looks at two different methods of fitting linear classifiers. The linear discriminant analysis is implemented in the MASS
More informationFlexibility and Robustness of Hierarchical Fuzzy Signature Structures with Perturbed Input Data
Flexibility and Robustness of Hierarchical Fuzzy Signature Structures with Perturbed Input Data B. Sumudu U. Mendis Department of Computer Science The Australian National University Canberra, ACT 0200,
More informationData Mining Practical Machine Learning Tools and Techniques
Output: Knowledge representation Data Mining Practical Machine Learning Tools and Techniques Slides for Chapter of Data Mining by I. H. Witten and E. Frank Decision tables Decision trees Decision rules
More informationComparative Analysis of Swarm Intelligence Techniques for Data Classification
Int'l Conf. Artificial Intelligence ICAI'17 3 Comparative Analysis of Swarm Intelligence Techniques for Data Classification A. Ashray Bhandare and B. Devinder Kaur Department of EECS, The University of
More informationSupport Vector Machines - Supplement
Support Vector Machines - Supplement Prof. Dan A. Simovici UMB 1 / 1 Outline 2 / 1 Building an SVM Classifier for the Iris data set Data Set Description Attribute Information: sepal length in cm sepal
More informationCyber attack detection using decision tree approach
Cyber attack detection using decision tree approach Amit Shinde Department of Industrial Engineering, Arizona State University,Tempe, AZ, USA {amit.shinde@asu.edu} In this information age, information
More informationBL5229: Data Analysis with Matlab Lab: Learning: Clustering
BL5229: Data Analysis with Matlab Lab: Learning: Clustering The following hands-on exercises were designed to teach you step by step how to perform and understand various clustering algorithm. We will
More informationMachine Learning: Algorithms and Applications Mockup Examination
Machine Learning: Algorithms and Applications Mockup Examination 14 May 2012 FIRST NAME STUDENT NUMBER LAST NAME SIGNATURE Instructions for students Write First Name, Last Name, Student Number and Signature
More informationOptimal Clustering and Statistical Identification of Defective ICs using I DDQ Testing
Optimal Clustering and Statistical Identification of Defective ICs using I DDQ Testing A. Rao +, A.P. Jayasumana * and Y.K. Malaiya* *Colorado State University, Fort Collins, CO 8523 + PalmChip Corporation,
More informationBasic Concepts Weka Workbench and its terminology
Changelog: 14 Oct, 30 Oct Basic Concepts Weka Workbench and its terminology Lecture Part Outline Concepts, instances, attributes How to prepare the input: ARFF, attributes, missing values, getting to know
More informationA Neural Network for Real-Time Signal Processing
248 MalkofT A Neural Network for Real-Time Signal Processing Donald B. Malkoff General Electric / Advanced Technology Laboratories Moorestown Corporate Center Building 145-2, Route 38 Moorestown, NJ 08057
More informationIntroduction to R and Statistical Data Analysis
Microarray Center Introduction to R and Statistical Data Analysis PART II Petr Nazarov petr.nazarov@crp-sante.lu 22-11-2010 OUTLINE PART II Descriptive statistics in R (8) sum, mean, median, sd, var, cor,
More informationMultiple Classifier Fusion using k-nearest Localized Templates
Multiple Classifier Fusion using k-nearest Localized Templates Jun-Ki Min and Sung-Bae Cho Department of Computer Science, Yonsei University Biometrics Engineering Research Center 134 Shinchon-dong, Sudaemoon-ku,
More informationBack propagation Algorithm:
Network Neural: A neural network is a class of computing system. They are created from very simple processing nodes formed into a network. They are inspired by the way that biological systems such as the
More informationA Self-Organizing Binary System*
212 1959 PROCEEDINGS OF THE EASTERN JOINT COMPUTER CONFERENCE A Self-Organizing Binary System* RICHARD L. MATTSONt INTRODUCTION ANY STIMULUS to a system such as described in this paper can be coded into
More informationFUZZY KERNEL K-MEDOIDS ALGORITHM FOR MULTICLASS MULTIDIMENSIONAL DATA CLASSIFICATION
FUZZY KERNEL K-MEDOIDS ALGORITHM FOR MULTICLASS MULTIDIMENSIONAL DATA CLASSIFICATION 1 ZUHERMAN RUSTAM, 2 AINI SURI TALITA 1 Senior Lecturer, Department of Mathematics, Faculty of Mathematics and Natural
More informationPrototype Selection for Handwritten Connected Digits Classification
2009 0th International Conference on Document Analysis and Recognition Prototype Selection for Handwritten Connected Digits Classification Cristiano de Santana Pereira and George D. C. Cavalcanti 2 Federal
More informationThe results section of a clinicaltrials.gov file is divided into discrete parts, each of which includes nested series of data entry screens.
OVERVIEW The ClinicalTrials.gov Protocol Registration System (PRS) is a web-based tool developed for submitting clinical trials information to ClinicalTrials.gov. This document provides step-by-step instructions
More informationHandwritten Hindi Numerals Recognition System
CS365 Project Report Handwritten Hindi Numerals Recognition System Submitted by: Akarshan Sarkar Kritika Singh Project Mentor: Prof. Amitabha Mukerjee 1 Abstract In this project, we consider the problem
More informationESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN ,
Interpretation and Comparison of Multidimensional Data Partitions Esa Alhoniemi and Olli Simula Neural Networks Research Centre Helsinki University of Technology P. O.Box 5400 FIN-02015 HUT, Finland esa.alhoniemi@hut.fi
More informationUse of Artificial Neural Networks to Investigate the Surface Roughness in CNC Milling Machine
Use of Artificial Neural Networks to Investigate the Surface Roughness in CNC Milling Machine M. Vijay Kumar Reddy 1 1 Department of Mechanical Engineering, Annamacharya Institute of Technology and Sciences,
More information